As PCIe versions strive for higher bandwidth and faster transfer rates, the increase in speed indirectly leads to excess heat. Intel is addressing this challenge by updating Linux drivers in a novel way. The open-source "PCIe Bandwidth Controller" is designed to automatically control link speed when heat reaches a certain threshold.
The core issue lies in the increasing speed requirements of PCIe, which demand better signal integrity and lower signal loss. Achieving this often necessitates higher clocks, increased power, and encoding optimizations, all of which generate additional heat. Take PCIe 5.0 solid-state drives, for instance. With a transfer rate of 32GT/s (twice that of PCIe 4.0), they frequently require dedicated cooling solutions to dissipate heat under sustained loads.
Intel's driver has been in development for over a year now. When temperatures rise too high, Linux can selectively reduce the PCIe link speed. It registers a "cooling device" state for each PCIe port, allowing for link speed adjustments. Lowering the PCIe link speed helps alleviate the thermal stress brought on by GPUs, SSDs, and other peripherals running hot.
In the latest patch notes, Intel engineer Ilpo Järvinen stated, "This series only adds support for controlling the PCIe link speed." While controlling the PCIe link width could also be useful, there doesn't seem to be a mechanism for this before PCIe 6.0 (L0p), so this series does not add link width throttling functionality.
As evident, the inability to adjust link width remains an unresolved limitation, but Intel is actively monitoring and planning to implement this feature.
Currently, the new Linux driver focuses solely on link speed throttling to mitigate heat dissipation issues. This straightforward software solution addresses the heat concerns arising from the ever-increasing PCIe speeds. By automatically reducing the link rate for each port, it offers a way to dynamically optimize cooling under load.
While initially seeming geared towards servers, this technology has proven beneficial for desktops and mobile systems as well. For example, if temperatures run too high, an x16 GPU link can temporarily drop to x8 mode. Of course, this means reduced bandwidth, potentially impacting hardware performance.
Last year, the PCI-SIG consortium released the first 0.3 draft review specification for PCIe 7.0 to its members. Though details are scarce, this announcement confirms that PCIe's development is progressing as planned. 7.0 will double the bandwidth of 6.0, achieving an astounding 512GB/s throughput via x16 links. It's not hard to imagine that the forthcoming standard will demand even stronger cooling capabilities.
With Intel engineers continuously refining the implementation of thermal throttling technology, the new PCIe bandwidth controller driver has undergone five revisions to date. While not yet finalized, adoption is likely around next year.
